how many cameras w/o AA filter (except MFDB) were in 2003 ?
Don't know exactly, but they have always been available in research and for astronomy, and by modifying existing sensors.
Besides that, not accounting for the difference in optical path (it's like using different apertures, or defocusing one lens) when trying to proclaim a benefit, is worse than sloppy science/research. Given the spirit at the time, it was plainly oriented at misleading people, and it apparently works till this day ...
The Hubel paper is presented looking as a research paper, but it was in fact a promotional piece by the stakeholder (which is fine, because they would have been grilled by the scientific community had they tried that route). They deliberately
compared apples to oranges in an attempt to bridge the gap in megapixels (X3) that the camera produced. The Raw converter even had a 2x(?) upsample setting.
They also were doing cycles/mm SFR tests, conveniently not mentioning that the 10D sensor had some 35% more sensels per image height and would therefore require 35% less output magnification than the SD10 for same size output, thus reducing the benefits in the SD10 scores. That was another omission to make the Foveon technology look better versus competition. They also made sure to compare (MTF) response (vertical axis) at a given cy/mm (which always benefits a sensor without AA-filter) and not real resolution (horizontal axis) as defined in the ISO 12233 standard they used incompletely by not normalizing for sensor size nor sensor pitch.
Don't get me wrong, eliminating the risk of false color aliasing is a huge benefit, but don't think that therefore there is no aliasing at all (another false claim). There is, it's luminance aliasing, and is there because it's unavoidable in discretely sampled imaging (unless truly diffraction limited by optics). It's always there, so any claim that there isn't any is a red flag
(!). It may be mild enough to be tolerable, but it is always there.
The luminance aliasing is much stronger without the use of an OLPF, but more tolerable in many cases because the luminance aliasing may sometimes look a bit like real detail, or look sharper than actual edges (like having built-in sharpening), and the overall MTF response is boosted which gives the images more punch. The drawback is that it also exaggerates stairstepping/jaggies on sharp high contrast edges and lines and produces false representations of surface structures. That can look a bit unnatural for the careful observer, but it's mainly visible in the exact plane of focus, so it might be remedied by shifting the focus a bit. These things are good to know in advance instead of finding out after the fact, which is why it helps to not ostracize the issues. Talking about them also allows to find solutions.